Retail has reached a tipping point, pivoting from speculative Generative AI pilots to the era of ‘agentic commerce. For those in executive roles, the central hurdle is no longer merely the adoption of AI, but the governance of its proliferation at a global scale.*

As organizations move toward a future defined by the operation of millions—and eventually billions—of autonomous agents, the infrastructure required to support these digital actors demands a fundamental re-architecture of the enterprise data stack and the underlying network fabric.*

The industry is currently navigating a transformation point where linear value chains are being replaced by autonomous, continuously orchestrated ecosystems.* Leading retailers are no longer viewing AI as a peripheral tool but as a core operating layer that bridges the gap between digital intent and physical fulfillment.* This evolution is underpinned by Google’s breakthroughs in megascale networking and AI-native data services, specifically the Virgo Network and the Agentic Data Cloud, which provide the high-concurrency inference and semantic context necessary for agents to function as reliable, high-fidelity extensions of the brand.

From Experimentation to Agentic Scale in Retail

By early 2026, the retail sector has reached a critical juncture in its technological maturation. Research from Deloitte indicates that 96% of global retail executives expect revenue growth, reflecting the belief that the “experimental phase” of AI is now behind the industry.

Industry forecasts suggest that by 2030, as much as 25% of global e-commerce sales may be influenced or handled autonomously by agents.* In the immediate term, nine in ten retail executives expect AI to become more prevalent than traditional search engines in 2026, and half expect the current multi-step shopping journey to collapse into a single-step conversational transaction by 2027.* This “collapse” represents the birth of agentic commerce, where the boundary between research and purchase is eliminated by autonomous systems capable of negotiating, deciding, and executing end-to-end economic transactions.

For retailers, agentic systems are projected to drive a 30% reduction in customer service operational costs by 2029, while 89% of retailers already report that AI initiatives are positively impacting revenue.*

To support this new shopping journey, the underlying hardware must evolve too.

Infrastructure of the Agentic Era: Virgo Network and the AI Hypercomputer

As the number of active agents in the retail ecosystem scales toward the billions, legacy data center architectures are proving insufficient. The “billion-agent future” requires an infrastructure that can handle massive-scale AI inference with near-zero latency and high resilience.

Google’s response to this challenge is the AI Hypercomputer, a system in which “compute” is no longer defined by the individual chip but by the entire data center fabric.

The Virgo Network: Megascale Data Center Fabric

The cornerstone of this new infrastructure is the Virgo megascale data center fabric. Designed specifically for the massive “east-west” traffic of modern AI training, Virgo provides the interconnectivity required for clusters reaching up to 134,000 TPUs or 80,000 GPUs in a single domain. Unlike traditional hierarchical designs, Virgo utilizes a multi-planar network architecture to provide up to 47 petabits/sec of non-blocking bandwidth.

Core technical pillars of Virgo include:

• Disjoint Multi-Plane Fabric: The network is divided into independent planes, each acting as a standalone Clos fabric with its own data and control planes. This elimination of “fate sharing” ensures that a software bug or hardware failure in one plane cannot cascade to others, maintaining cluster-wide availability.
• Rapid Fault Mitigation: Virgo is engineered to handle “stragglers” (nodes slowed by congestion or faults). If a plane experiences a hardware fault, traffic is automatically diverted to healthy planes. This ensures mission-critical AI computations proceed without the synchronization stalls that often plague large-scale training.
• High-Radix, Low-Latency Topology: By leveraging high-radix switching and advanced cable management, Virgo employs a flat, two-layer non-blocking topology. By collapsing the traditional three-stage Clos into two stages, Google has reduced unloaded fabric latency by 40%, significantly accelerating inter-accelerator communication.
• Power Efficiency via LPO: Virgo marks a shift toward Linear Pluggable Optics (LPO). By removing the power-intensive Digital Signal Processors (DSPs) from the optical modules, network power consumption is reduced by up to 30% per module. This allows data centers to reallocate megawatts of power from the network directly to compute workloads.
• Optical Circuit Switching (OCS): Virgo integrates Google’s proven OCS technology (Apollo), which allows the network to be reconfigured at the optical layer. This provides the flexibility to dynamically adjust the topology to match the specific communication patterns of different AI models.

For a retailer running millions of agents, Virgo provides the necessary “concurrency highway.” When agents perform complex reasoning tasks or “All-Reduce” operations across distributed datasets, Virgo’s ability to perceive congestion in real-time and dynamically adjust traffic ensures that agents do not experience the “jitter” that typically degrades performance in large-scale systems.

TPU 8i: Optimized for Agent-Scale Inference

Complementing the Virgo Network are Google’s eighth-generation Tensor Processing Units (TPUs), specifically the TPU 8i. While the TPU 8t is optimized for compute-intensive training, the TPU 8i is purpose-built for low-latency, high-throughput inference—the lifeblood of active agents.

The TPU 8i connects 1,152 TPUs in a single pod, offering three times as much on-chip SRAM as previous generations. This architecture is critical for Mixture-of-Experts (MoE) models, which are increasingly used to power sophisticated retail agents that need to switch between domain-specific “experts” (e.g., pricing, logistics, styling) in real time. By delivering an 80% improvement in performance per dollar, TPU 8i enables retailers to run millions of concurrent agents cost-effectively.*

Multi-Planar Resilience in AI Clusters

The shift to multi-planar networking is a direct response to the increasing cost of outages in an agentic world. As systems grow more complex, the financial consequences of downtime are amplified because agents are deeply embedded in revenue-generating workflows.

The reduction in the physical and logical distance between compute nodes is what allows an agent to respond to a customer mission—such as “plan a kitchen renovation within my budget”—by querying thousands of product variables and supply chain constraints in milliseconds. The mathematical optimization of these networks centers on minimizing tail latency across the fabric.

The Agentic Data Cloud: Turning Data into Context

For agents to act as trusted representatives of a retail brand, they require more than just raw data; they require “enterprise context”.* The Agentic Data Cloud is Google’s unified architecture designed to turn fragmented, siloed data into a shared intelligence layer that agents can reason over reliably at scale.

From Passive Repositories to Systems of Action

Traditionally, retail data platforms were “systems of intelligence”—passive repositories where humans would run queries to understand what happened in the past. The Agentic Data Cloud transitions these into “systems of action,” foundations that agents use to sense signals, make decisions, and execute transactions autonomously. This re-architecture is built on several key components:

• Knowledge Catalog: An evolution of the Dataplex Universal Catalog, it serves as a semantic core. It automatically enriches structured and unstructured data, extracting entities and relationships so agents understand business semantics.
• Smart Storage: Within Google Cloud Storage (GCS), Smart Storage automatically tags, generates embeddings, and extracts entities from unstructured data as soon as it lands.
• Data Agent Kit (Preview): A set of tools that allows developers to build “data-aware” agents that interact with governed datasets and apply business logic across systems.

The Role of Semantics and Grounded AI

A common failure point in retail AI pilots is the “data graveyard”—vast amounts of dark data that agents cannot interpret. The Agentic Data Cloud solves this by creating a “unified semantic layer”. By standardizing on open-table formats like Apache Iceberg, Google enables bi-directional federation across clouds, allowing agents to query data wherever it lives.

This grounded context is essential for accuracy. When a retailer like Kingfisher uses Vertex AI Search for Commerce, the agents are grounded in the company’s extensive product and data catalogs through the Knowledge Catalog. This ensures that when an agent suggests a specific tool for a DIY project, it has verified availability, compatibility, and the customer’s loyalty status.

Memory and Personalization at Scale

Effective agentic personalization requires long-term memory. Google has introduced “Memory Bank” and “Memory Profiles” to provide agents with persistent context over time.* Unlike temporary sessions, these features allow an agent to recall user-specific details, project histories, and preferences across months of interactions.

Infrastructure Component	Data Function	Agent Utility
Agent Sessions	Transactional State	Tracking real-time context and session IDs
Memory Bank	Persistent Context	Recalling user preferences and constraints over time
Knowledge Catalog	Semantic Core	Governing business rules and relationship mapping
Smart Storage	Autonomous Enrichment	Automated tagging and embedding of dark data

Redefining the Customer Mission: Personalization 2.0

The paradigm of retail discovery is shifting from “products” to “missions”. Customers are increasingly looking to solve complex problems, such as “refresh my winter wardrobe”, rather than browsing for individual SKUs. Agentic AI is the primary vehicle for this shift, providing the interface that best grasps the customer’s context, constraints, and preferences.*

The “Bazaar” vs. “Walled Garden” Futures

BCG identifies several potential agentic futures for the retail sector:

• The Open Agentic Bazaar: Shopping agents browse and transact freely across brands. Retailers evolve into “network hubs,” providing rich product data and real-time inventory.
• Brand Resurgence Through Data Fortresses: Large platforms dominate by uniting search, commerce, and their own agent engines in closed “walled gardens.” Success depends on owning the proprietary agentic layer.

Case Study: Kingfisher’s Move to Agentic Commerce

Kingfisher recently announced a multi-year partnership with Google Cloud to lead the era of agentic commerce across the UK and Europe. By moving beyond traditional keyword search toward Vertex AI Search for Commerce, Kingfisher is enabling proactive, AI-driven shopping assistants.

These agents empower customers to:

• Plan Complex Projects: Agents generate tailored shopping lists for multi-stage DIY tasks.
• Conversational Discovery: Shoppers interact with agents intuitively, reflecting how they think about home improvement.
• Seamless Execution: Agents execute purchases directly, integrated with Kingfisher’s central data lake, Nucleus.

Autonomous Logistics and Payments: The “Last Mile” of Agency

The agentic revolution is not confined to the digital storefront; it extends into physical logistics and financial payments. Physical AI”—the use of AI to control robots and machines—is transforming production and logistics, providing a competitive advantage for those who integrate it with digital agents.*

Agentic Payments: Nexi and the AP2/UCP Protocols

A critical barrier to fully autonomous commerce is an agent’s ability to execute a secure, authorized payment. Nexi Group has partnered with Google Cloud to build the foundational infrastructure for agentic payments in Europe. This partnership centers on two key protocols:

• Universal Commerce Protocol (UCP): An open-source standard to orchestrate the end-to-end AI commerce lifecycle
• Agent Payments Protocol (AP2): A specialized trust layer that enables authorized, secure payments through cryptographically signed mandates

Physical AI and Warehouse Automation

In the warehouse, physical AI is the “silent revolution” of 2026. EY identifies physical AI as a major trend, with AI-controlled robotics optimized for logistics. Lush, for instance, uses Google Cloud’s Vertex AI to power “Lush Lens,” which identifies packaging-free products without barcodes, drastically reducing checkout times and improving inventory management.*

The Organizational Blueprint for the Billion-Agent Era

Achieving the full potential of the Billion-Agent Future requires a radical simplification of processes across the enterprise. McKinsey research highlights that value from AI depends as much on people as technology, recommending that for every $1 spent on technology, $5 should be spent on people and capability building.

Workflow Redesign and Human-Agent Collaboration

The default workforce model is transitioning to “human-agent teams”. One in four leaders expect AI agents to act as autonomous teammates in the short term.* To prepare, retailers must invest in “AI fluency”—equipping teams with the skills to oversee agentic outputs and apply human critical thinking where judgment is paramount.*

Strategic Steps for Executives

1. Detect High-Value Processes: Identify 2-3 end-to-end processes for agentic deployment while applying “kill criteria” to experiments that fail to scale.
2. Establish the Control Plane: Manage a fleet of agents using Agent Identity and Registry tools to ensure discoverability and governance.
3. Scale with a Unified Stack: Standardize on an integrated stack of chips (TPU 8i), models (Gemini), and data (Agentic Data Cloud) to avoid fragmented infrastructure.

Orchestrating the Transformation

The transition to an agentic future is no longer a question of “if,” but of “how quickly.” For retail executives, the period between 2024 and 2026 has provided the foundational blocks—megascale networking with Virgo, semantic data layers with the Agentic Data Cloud, and secure payment protocols like AP2—to build a truly autonomous enterprise.

However, the path to increased new demand and significant operational cost reductions lies in organizational rewiring. The retailers who will lead in 2027 and beyond are those who have embraced agentic AI as the new operating layer of their business.

As a Premier Google Cloud Partner, Kartaca is uniquely positioned to help retailers navigate this complex transition. Our expertise ensures that your agentic workforce is not only powerful and scalable but also governed, compliant, and deeply integrated into your core retail operations.

Ready to start your journey toward agentic commerce? Contact us today to explore our specialized AI transformation services for modern retailers and orchestrate your global personalization strategy.